Risk prediction manuscript

Plot Survival vs Guan score

Risk prediction summary by quality of model

Summary statistics by subtype

true and false positive percentages

	TPR			FPR
subtype	low	high/extreme	extreme	low	high/extreme	extreme
FGFR3	88.0	86.0	80.0	12.00	14.0	20.00
t(4;14)	87.5	85.7	90.9	12.50	14.3	9.09
del1p	97.0	36.0	29.0	3.00	64.0	71.00
amp1q	94.0	64.0	53.0	6.00	36.0	47.00
del13	98.0	46.0	54.0	2.00	54.0	46.00
agnostic	92.5	31.8	19.7	7.45	68.2	80.30

SYGNAL risk prediction for each subject was chosen based on the quality of the subtypes each subject exhibited. If a subject exhibited multiple subtypes then the average of the best quality models risk prediction was used to classify the subject. If no subtypes were exhibited then the “all” model was used. For example, if a subject exhibited the t(4;14), amp1q, and del(1p) subtypes then the t(4;14) subdtype risk prediction would be used for that subject.

subtype	AUC	quality
FGFR3	0.932	A
t(4;14)	0.918	A
del(13)	0.890	B
del(1p)	0.878	B
amp(1q)	0.843	B
agnostic	0.724	C

subtype	AUC	sky92	gep70	n	quality
FGFR3	0.932	0.661	0.779	77	A
t(4;14)	0.918	0.670	0.745	105	A
del(1p)	0.890	0.614	0.616	60	B
amp(1q)	0.878	0.655	0.736	203	B
del(13)	0.843	0.674	0.737	190	B
agnostic	0.724	0.615	0.648	769	C

IA12

Scatter plots by subtype

ia12 PFS cors

subtype	mmSYGNAL	SKY92	GEP70
t(4;14)	-0.068	-0.243	-0.158
FGFR3	0.069	-0.255	-0.123
amp(1q)	-0.166	-0.219	-0.350
del(1p)	-0.077	-0.188	-0.181
del(13)	-0.057	-0.291	-0.307
agnostic	-0.056	-0.293	-0.244

ia12 Guan cors

subtype	mmSYGNAL	SKY92	GEP70
t(4;14)	0.332	0.261	0.304
FGFR3	0.500	0.335	0.490
amp(1q)	0.597	0.280	0.469
del(1p)	0.541	0.262	0.448
del(13)	0.648	0.301	0.559
agnostic	0.624	0.372	0.506

Full model: highest quality

50% survival probability PFS (months)

methods	extreme	high	low
SYGNAL	18.4	27.4	39.9
GEP70	NA	20.0	40.5
SKY92	NA	15.8	39.1

logrank FDR values

risk	mmSYGNAL	GEP70	SKY92
extreme	NA	NA	NA
high	0	0	0

t(4;14)

50% survival probability PFS (months)

methods	extreme	high	low
SYGNAL	20	22.1	45.2
GEP70	NA	13.1	29.6
SKY92	NA	12.4	29.6

logrank FDR values

risk	mmSYGNAL	GEP70	SKY92
extreme	NA	NA	NA
high	0	0	0

FGFR3

50% survival probability PFS (months)

methods	extreme	high	low
SYGNAL	20.4	23.6	59.0
GEP70	NA	13.1	29.0
SKY92	NA	12.4	28.2

logrank FDR values

risk	mmSYGNAL	GEP70	SKY92
extreme	NA	NA	NA
high	0	0	0

del(1p)

50% survival probability PFS (months)

methods	extreme	high	low
SYGNAL	9.915	5.9	NA
GEP70	NA	13.0	NA
SKY92	NA	9.7	NA

logrank FDR values

risk	mmSYGNAL	GEP70	SKY92
extreme	NA	NA	NA
high	0	0	0

del(13)

50% survival probability PFS (months)

methods	extreme	high	low
SYGNAL	11.4	21.3	44.3
GEP70	NA	20.0	32.6
SKY92	NA	15.0	31.7

logrank FDR values

risk	mmSYGNAL	GEP70	SKY92
extreme	NA	NA	NA
high	0	0	0

amp(1q)

50% survival probability PFS (months)

methods	extreme	high	low
SYGNAL	15	20.8	45.2
GEP70	NA	16.6	42.0
SKY92	NA	18.4	39.2

logrank FDR values

risk	mmSYGNAL	GEP70	SKY92
extreme	NA	NA	NA
high	0	0	0

agnostic

50% survival probability PFS (months)

methods	extreme	high	low
SYGNAL	11.3	23.9	41.3
GEP70	NA	20.0	40.5
SKY92	NA	15.8	39.1

logrank FDR values

risk	mmSYGNAL	GEP70	SKY92
extreme	NA	NA	NA
high	0	0	0

Subtype specific models

Quality models

50% survival probability PFS (months)

quality	extreme	high	low
A	20.0	20.8	45.2
B	14.2	23.6	45.2
C	11.3	23.9	41.3

	50% survival probability PFS			n
quality	extreme	high	low	extreme	high	low
A	20.0	20.8	45.2	37	10	59
B	14.2	23.6	45.2	37	31	257
C	11.3	23.9	41.3	37	90	642

Best quality model

GSE19784

Scatter plots by subtype

GSE19784 Guan cors

method	correlation
mmSYGNAL	0.276
SKY92	0.537
GEP70	0.388

GSE19784 PFS cors

method	correlation
mmSYGNAL	-0.275
SKY92	-0.529
GEP70	-0.379

Full model: all subjects

Full model ROC (first plot) uses all subjects with SYGNAL using the best quality model.

Med PFS 50% (months)

methods	extreme	high	low
SYGNAL	8.845	20.740	31.517
GEP70	NA	6.253	30.602
SKY92	NA	3.355	28.873

logrank FDR values

risk	mmSYGNAL	GEP70	SKY92
extreme	NA	NA	NA
high	0	0	0

t(4;14)

t(4;14) ROC curve only shows the risk prediction for subjects exhibiting the t(4;14) subtype.

Med PFS 50% (months)

methods	extreme	high	low
SYGNAL	14.03	4.982	NA
GEP70	NA	7.930	30.093
SKY92	NA	7.930	30.093

logrank FDR values

risk	mmSYGNAL	GEP70	SKY92
extreme	NA	NA	NA
high	0	0	0

GSE24080

Scatter plots by subtype

GSE24080 Guan cors

method	correlation
mmSYGNAL	0.332
SKY92	0.335
GEP70	0.338

GSE24080 PFS cors

method	correlation
mmSYGNAL	-0.287
SKY92	-0.288
GEP70	-0.371

Compare Guan to survival clinical outcome

Risk stratification by genetic abnormality subtype counts

Risk stratification by genetic abnormality subtype was done by first calculating the number of high risk subtypes (del(17p), 1q+, FGFR3, WHSC1 and MAF) each subject showed. All subjects were then ranked by number of subtypes (0 to 6). For each subtype count category, subjects were randomly ordered.

Rank ordering within each subtype count category was based on SYGNAL risk probability for the stratification that combined SYGNAL risk probability and subtype stratification.

Cytogenetics: change risk distribution

IA12 cytogenetics

cyto count	total	high	low
0	456	35%	65%
>0	313	40%	60%
1	194	38%	62%
2	81	40%	60%
3	33	55%	45%
4	5	40%	60%

Sample sizes of number of mutations

mutations	counts
4	5
3	33
2	81
1	194
0	456

KM survival plots

Risk stratification by subtype abnormality counts

del(13) and MAF subtypes are not included in subtype abnormality counts.

Table of log rank test for difference of survival curves for subjects with 0 abnormalities vs those with 1 to 5 abnormalities.

log rank p-values

count	p
1	0.660
2	0.281
3	0.033
4	0.448
0 v 1+	0.170

50% survival probability PFS (months)

abnormalities	50% PFS	n
4	12.7	5
3	23.8	33
2	30.1	81
1	32.6	194
0	39.1	456

KM plots cyto vs mmSYGNAL

Full model: all subjects

Full model ROC (first plot) uses all subjects with SYGNAL using the best quality model.

50% survival probability PFS (months)

methods	extreme	high	low
SYGNAL	12.064	28.653	74.149
GEP70	NA	20.842	74.149
SKY92	NA	22.333	74.589

logrank FDR values

risk	mmSYGNAL	GEP70	SKY92
extreme	NA	NA	NA
high	0	0	0

t(4;14)

t(4;14) ROC curve only shows the risk prediction for subjects exhibiting the t(4;14) subtype.

50% survival probability PFS (months)

methods	extreme	high	low
SYGNAL	12.268	31.042	NA
GEP70	NA	16.063	NA
SKY92	NA	13.657	64.084

logrank FDR values

risk	mmSYGNAL	GEP70	SKY92
extreme	NA	NA	NA
high	0	0	0

Full plots

ROC

Highest quality model

t(4;14)

KM

Highest quality model

highest quality model

	IA12			GSE19784			GSE24080
methods	extreme	high	low	extreme	high	low	extreme	high	low
SYGNAL	18.4	27.4	39.9	8.85	20.70	31.5	12.1	28.7	74.1
GEP70	NA	20.0	40.5	NA	6.25	30.6	NA	20.8	74.1
SKY92	NA	15.8	39.1	NA	3.36	28.9	NA	22.3	74.6

	IA12			GSE19784		GSE24080
methods	extreme	low	high	low	high	low	high
mmSYGNAL	628	68	73	233	49	474	85
GEP70	608	0	161	205	77	503	56
SKY92	673	0	96	278	4	547	12

t(4;14)

t(4;14)

	IA12			GSE19784			GSE24080
methods	extreme	high	low	extreme	high	low	extreme	high	low
SYGNAL	20	22.1	45.2	14.03	4.982	NA	12.268	31.042	NA
GEP70	NA	13.1	29.6	NA	7.930	30.093	NA	16.063	NA
SKY92	NA	12.4	29.6	NA	7.930	30.093	NA	13.657	64.084

Summary: predicted risk class

risk		IA12			GSE19784			GSE24080
clinical	predicted	mmSYGNAL	GEP70	SKY92	mmSYGNAL	GEP70	SKY92	mmSYGNAL	GEP70	SKY92
extreme	extreme	17	0	0	4	0	0	8	0	0
extreme	high	8	29	23	8	12	8	5	15	18
extreme	low	46	42	48	28	28	32	30	28	25
high	extreme	33	0	0	7	0	0	9	0	0
high	high	26	65	33	21	9	3	28	35	40
high	low	156	150	182	94	113	119	75	77	72
low	extreme	18	0	0	0	0	0	4	0	0
low	high	39	67	40	9	1	0	31	27	41
low	low	426	416	443	111	119	120	369	377	363

method	IA12	GSE19784	GSE24080
mmSYGNAL	0	0	0
GEP70	0	0	0
SKY92	0	0	0

ROC mmSYGNAL